Inventing a template for educational facilities to optimize circulation, daylighting within various climates around the world
Jonathan Chester, Hao-Yeh Lu, Brian Turner, Daniel Joshua Vanderhorst, Andy Chia-Jun Wen,
Many generative design applications aim to optimize space planning variations in plan or in elevation separately. We are considering if you can put these two tasks into one. SANAA’s Zollverein School of Management and Design is a cubic structure with a series of seemingly random square openings. We are considering if there is a connection between the floor plan and the elevation. Our design problem involves how to design this cubic volume, while considering both the plan efficiency and the rationalization of the elevation.
Parametric designs struggle to advance beyond the bifurcation of optimizing plans (for circulation) or elevation (for energy and lighting). This separation is often disregarded in offices because of the lengthy design processes and the ample amounts of labor that allow for a slow integration of work. The challenge lies in the inefficiencies that come with this disconnect in the process. Design approaches that lack optimization remain slow and inefficient compared to newer generative techniques. Yet within the newer systems, the amount of time and energy lost in the back and forth effort to integrate separately optimized design creations of plan and section would be better spent on other characteristics of the project. Additionally, these inefficiencies may force unnecessary surrender of the very things that were meant to be enhanced (such as sustainability, efficiency, and overall good design).
Fig 2. SANAA, 2010, Zollverein School of Management & Design
Our project goals can be articulated as the effective integration of plan and elevation optimization, while retaining the most successful performative benchmarks of core circulation, energy consumption, and programmatic daylighting that provide enjoyable and easily navigable spaces for users that are sustainable and economical. Our methodology allows for a seamless process for creating.
Fig. 3.0.1. Overall Workflow of generative optimization process including inputs, model, discover, and outputs.
3.1 Design Space Model
3.1.1 Plan Reorganization
We begin the process by establishing the base of our design from the 3 integral cores within the precedent project. The existing grid of 6’x 6’ is reused to distribute the cores by a set of logics that guide the project.
These cores are distributed throughout the plan based on two principles: maximum coverage and avoiding adjacencies.
The first principle of maximum coverage requires the first two established cores to cover a determined distance over most of the plan. In order to achieve that, we use Discover Categorical Input to randomly pick the first core, and pick the second core from one of the locations in a certain minimum distance.
Fig. 220.127.116.11. Cores distribution principle: maximum coverage
The second principle of avoiding adjacencies creates a smaller distance around the two main cores preventing any designs where the third and smallest core is directly attached to it. This ensures that the cores are spaced a certain distance for optimum circulation, structural composition, and away from the valuable light and views of the exterior walls.
Fig. 18.104.22.168. Cores distribution principle: avoiding adjacencies
After determining the core layout, we use Discover to pick an X and Y axis based on the edge of the main cores.
Fig. 22.214.171.124. Picking axis as the space splitting gene.
From these axes our rooms are distributed via genetic algorithms. A predetermined list of rooms and their areas was created from the precedent. From there, each individual floor is split by the XY axis created by the core.
The resulting spaces split by the axis create a hierarchy of spaces within each floor. These spaces are assigned an individual room, matching each based on largest to smallest square footage.
Using simple math, we combine the difference between each split space and its assigned target room, and count the squareness of each room. It creates a score per generated design, allowing for the removal of room designs with negative scores that can’t accommodate the target.
Fig. 126.96.36.199. Optimizing floor plan
Additionally, corridors are disregarded as minimal spaces that can be assigned after the main programs have been distributed and optimized.
Fig. 188.8.131.52. Hundreds of generated models within Discover
3.1.2. Maximization Elevation Ratio of Opening
Previously in the plan, we aimed to give each space a certain ratio of openness and adjacencies to the facade, followed by finding creating an efficiency of circulation.
Our next task is to panelize the entire facade so that each facade ‘cell’ could be assigned a value based on the recommended light level of the room within. Each of these cells could then be given a prefabricated facade panel that has a number of perforations of glazing corresponding to the aforementioned light level. Next, the elevation is derived from the optimized core and programs to create both efficient and aesthetically interesting facades based on each room per floor.
Our process starts by dividing up the façade of the cubic building into equal horizontal lengths of 7m (floor to floor heights) and a vertical grid which determines the width of the facade panels. The width of this grid was determined based on the grid that the plan was based on: a 6 x 6 grid of 5.4m with a 1.8m perimeter around it. This resulted in each floor having 24 5.4m x 7m facade panels.
Ten different panel types are created using three different window sizes. The placement of the windows is loosely based on the same aesthetic facade organization of the original SANAA building. Different window arrangements and permutations are used to allow the panels to allow in different amounts of light. These are organized on a scale of 1 to 5: 1 being the panel that has no windows and thus allows in no natural daylight and 5 being the option with the most window surface area. The more perforated panel options — those fitting into the 5 category — would be the most suitable for the rooms requiring the most natural lighting. While there are only ten panel types, these can be installed upside down and mirrored, thus allowing the 10 rectangular pre-fab panels to be installed in almost 40 different orientations and create maximum facade diversity. A list of lighting levels per program was taken from an exterior source.
After the optimization of the core and program has placed all the rooms, and all the panel logic is set, the rooms of different light levels are averaged with the other rooms they touch, allowing for a seamless facade that prevents a patchwork look for the windows. The randomization of panels is taken from a group that matches the required 1–5 lux value we established. This also allows for minimal panel repetition and allows the generation of the remaining panels until all of the ten are used up.
To showcase the process for distributing the panels, the process starts with calculating the daylight that hits the facade, based on sun orientation, to determine the light level at the point. The existing panel list is channeled into a randomizer tool in grasshopper that picks one of the ten panels within a certain type, such as a light level of 2, and then places that panel, creating a variety on the facade. In aggregate, the whole building will look different per each generated run.
Fig. 184.108.40.206. Panel Lux Sampling
Fig. 220.127.116.11 Facade Panel Types Options
Fig. 18.104.22.168 Facade Panel Solar Studies for a singular design Iteration (from left to right — spring, summer, fall, winter)
3.2 Input Parameters
- Core location and orientation: Discover Categorical Input
- Gene Grid Selection: Discover Categorical Input
- Space subdivide parameter: Discover Continual Input
3.3 Output Performance Metrics: Daylight Simulation
Based on the optimized plan partitions, combined with the facade panel options decided by the daylight requirement score, which could generate the whole buildings. Furthermore, we add on the orientation of the building as a variable. Through the approach of Honey Bee Annual Daylight Metrics Analysis, those different orientations could be evaluated with different Daylight Scores correspondingly.
With the optimization of cores, plans, and facades, now we can use daylighting analysis as an objective to further determine and confirm that maximum daylighting has been used to establish the generative process. For this process, we made ample use of the input data.
Fig 3.3.1 Honey Bee Annual Daylight Metrics Analysis
Fig 3.3.2 Daylight Autonomy Simulation Honeybee Component
Based on the plan and facade optimization approach, we aim to maximize daylighting due to the accommodation of both plan and façade. Using general daylighting values, the scale and range are simplified to match the façade scoring of 5 levels of light. More specifically, we take the already established program’s recommended daylight lux levels, ranging from around 100 to 500 lux. From here, the average of the rooms can be taken and used subsequently. For example, a room whose average is setup to have a max of 300 lux can be concurrently used to generate a heat map based on the 5 established colors corresponding to the minimum and maximum lux levels. Each value per room will have its score of daylight based on each room’s centered average spatial point.
We can transfer this score to Discover and establish it as an objective component. From here, we try to maximize the sum of the plan per individual floor. We start with the first of the five floors. The different floors will be determined by the result. Once all the floors have been maximized per floor, then the total plan has been optimized for maximum necessary daylighting.
Fig 4.1 Daylight Autonomy Simulation Score
Fig 4.2 Summary video of generative design problem, process, and results.
It is apparent that the overall optimization resulted in our intended goal to explore a generative design process into more three-dimensional applications, considering not only the plan layout optimization but the sectional and elevational composition as well.
Throughout the optimization stages, we found that separating the different optimizations made the process go more smoothly. The initial idea was to consider the parameters of the plan and facade simultaneously. However, that workflow makes too many design options which might lead to bad options in plan and in facade. We ended up optimizing the plan according to its required program area first. Then we used the result of the optimized plan to generate possible facade designs. Our decision to do this helped us eliminate many design options that might not be worthy to take time analyzing.
This type of generative design would be particularly applicable in the design field if it could be expanded to different sizes and footprints. It could greatly help simplify the workflow for designing large projects — such as towers — that have much complexity in their design. Going forward, it would be beneficial to allow other shapes than square plans and allow for more floors in order to increase the applicability of this as a design tool.
Ultimately, we found the integration of the space planning and the elevation daylighting resulted in a more efficient design process that, with optimal energy consumption mitigation, can be applied to any climate and location in the world.
Fig 5.1. Solar study of selected optimized model for a New York City site